Throwing Light on PyTorch

The Pythonic Deep Learning Framework

The What and Why of PyTorch ?

Python based Deep Learning Framework
Developed by Facebook Research Group

Dynamic Computational Graphs
Easy Debugging
More Pythony

What ?

Why ?

Getting Started with PyTorch

Tensors : Lego Blocks of Neural Networks

Numpy Arrays on GPU Steroids

Tensors : Lego Blocks of Neural Networks

Numpy Conversion Bridge

Tensors : Lego Blocks of Neural Networks

Easy switch between CPU and GPU

Neural Networks

Forward Propagation

Neural Networks

Backpropagation

Neural Networks

Problem

Neural Networks

Solution : Computational Graphs

Static Computational Graphs

Define the graph structure beforehand
No Actual Data, Only placeholders
Used by Tensorflow

Dynamic Computational Graphs

Define the graph on the fly
Feed in Actual Data
Used by PyTorch

Simple Neural Network in PyTorch

AutoGrad (Automatic Differentiation)

AutoGrad Package is central to all Neural Networks in PyTorch.

It provides classes and functions implementing automatic differentiation of arbitrary scalar valued functions.

It requires minimal changes to the existing code.

Tensors Continued (Previously Variables)

It is a part of the AutoGrad Package.
It allows for computing the gradient.
Unlike TensorFlow tensors, PyTorch tensors have actual data.

Tensor Frame

Tensors Continued (Previously Variables)

Neural Networks in PyTorch

They are nothing but Computational Graphs.
Supports Dynamic Computational Graphs.
Varying level of abstraction according to usage and expertise.

Neural Networks in PyTorch

import torch.nn as nn

# Example of using Sequential
model = nn.Sequential(
          nn.Conv2d(1,20,5),
          nn.ReLU(),
          nn.Conv2d(20,64,5),
          nn.ReLU()
        )



# Example of using Sequential with OrderedDict
model = nn.Sequential(OrderedDict([
          ('conv1', nn.Conv2d(1,20,5)),
          ('relu1', nn.ReLU()),
          ('conv2', nn.Conv2d(20,64,5)),
          ('relu2', nn.ReLU())
        ]))

Keras of PyTorch

Neural Networks in PyTorch

class MnistModel(nn.Module):
    def __init__(self):
        super(MnistModel, self).__init__()

        # input is 28x28
        # padding = 2 for same padding
        self.conv1 = nn.Conv2d(1, 32, 5, padding=2)

        # feature map size is 14*14 by pooling
        # padding = 2 for same padding
        self.conv2 = nn.Conv2d(32, 64, 5, padding=2)

        # feature map size is 7*7 by pooling
        self.fc1 = nn.Linear(64*7*7, 1024)
        self.fc2 = nn.Linear(1024, 10)
        
    def forward(self, x):

        x = F.max_pool2d(F.relu(self.conv1(x)), 2)
        x = F.max_pool2d(F.relu(self.conv2(x)), 2)
        x = x.view(-1, 64*7*7)   
        x = F.relu(self.fc1(x))
        x = F.dropout(x, training = self.training)
        x = self.fc2(x)
        return x

Neural Networks in PyTorch

## Saving only model parameters.
torch.save(model.state_dict(), Path)

## Pickle entire model.
torch.save(model, Path)

## Loading model parameters.
torch.save(modelParamPath)

## Loading entire model.
torch.load(modelPath)

Saving and Loading Neural Nets in PyTorch

Questions ?

rahulbaboota

Throwing Light on PyTorch

By Rahul Baboota

Throwing Light on PyTorch

An introduction to Facebook AI's Deep Learning Library.

Rahul Baboota

Computer Science Junior